In this project we fine-tune the DisilBERT model and compare it to a baseline SVM classifier on the multiclass task of classifying pro-eating disorder users on Twitter.
To set up project using the correct Python packages, run this command
pip install -r requirements.txt
preprocess_tweets.ipynb imports the original dataset of 9.821.107 tweets and aggregates them into 6824 rows of text document, where each row represents one Twitter-user.
distilbert-finetune.py includes the fine-tuning of the DistilBERT-base Sequence Classifier.
DistilBERT_predictions.ipynb includes importing the trained DistilBERT model and predictions.
SVM_classifier.ipynb includes the training and testing of the SVM classifier